Aleph Alpha's 78B Kolibri model challenges US AI dominance
aleph alpha releases kolibri, a 78b german-english model, to promote european ai sovereignty. it outperforms comparable models on quality and cost.
aleph alpha released kolibri, a german-english language model with 78 billion parameters, 3 billion of which are active per token. it was trained on 768 b200 gpus in germany and finland and supports context windows of up to one million tokens. kolibri scores 71% on german benchmarks and outperforms comparable models like gpt oss a5b and qwen 3.6 a3b. the model was developed under european law with the eu ai act in mind.